Could a professional be negligent for not using AI?

Yes, in the considered view of the body that has looked hardest at the question. The UK Jurisdiction Taskforce’s legal statement of July 2026 says a professional “could be liable for failing to use AI in circumstances where a competent member of their profession would have done so”. The same statement says the professional can be liable for using it carelessly: for skipping due diligence on the tool, for not testing it, for not checking its output. The two duties are one duty, to exercise reasonable care and skill, and it now points in both directions. The standard is set by what competent peers do, so it moves as adoption moves. What it does not move is the requirement to remain able to check, and that is the part of the answer a profession controls.

Where the claim comes from#

The UK Jurisdiction Taskforce is an industry-led body that promotes English law for technology. In July 2026 it published its Legal Statement on Liability for AI Harms under the private law of England and Wales, drafted by a team led by Matthew Lavy KC, after a consultation that opened in January. The statement itself could not be retrieved for this page; it was read through the analyses of Legal IT Insider, Cooley, Mishcon de Reya, Burges Salmon and TLT. Lavy’s summary, as Legal IT Insider quotes it: “English law’s existing approach to legal liability already provides a coherent framework for analysing where responsibility should fall.” On professionals, Cooley’s account has the statement saying that “failure to use AI for a task when a professional exercising reasonable care and skill would have done so, can also lead to liability”. The wider statement, and what it says about who is liable when an agent acts, is on can a company blame its AI agent.

The standard has two edges#

The statement does not create a duty to adopt technology. It restates the ordinary standard: a professional must do what a reasonably competent member of the profession would do. What is new is that the answer now sometimes includes a tool. If competent peers use AI for a task and a client is harmed because a practitioner did not, the omission can be a breach. The other edge is sharper and older. Cooley lists the indicators of breach the statement names for those who do use AI: “failure to conduct proper due diligence on an AI system”, “failure to ensure there has been sufficient testing to check the AI system is suitable”, and “failure to exercise oversight of the AI system’s output”. Legal IT Insider adds that the statement expects professionals to check for hallucinations before relying on output. And Cooley records the statement’s hardest line: where a court concludes that the professional could not reasonably have done the due diligence, “the professional should not have used the AI tool at all”. A tool one cannot check is a tool one may not use. Nothing here binds a court; a legal statement is an analysis of what the law is likely to do, and no English judgment on either edge has been reported.

What competent peers are doing, measured#

Because the standard is set by peers, the first question for any profession is what its members have adopted. For one UK profession there is a measurement. Derksen and colleagues at Queen Mary University of London surveyed 598 UK general practitioners in autumn 2025 and reported in npj Digital Medicine in May 2026 that about 40 per cent were using AI scribes, with a further 23 per cent having used them, across a mean of 60 per cent of consultations among users. Over 75 per cent endorsed the timeliness benefits. Over 60 per cent agreed there are risks of inaccuracies, errors, misinterpretations and associated medicolegal threats. The sample over-represents younger and male GPs, so the proportions are indicative rather than national. Read against the statement, the figures carry a double message. A GP who declines a scribe is already in a minority and, if a court took the survey as the measure of competent practice, on the wrong side of the first edge. A GP who uses one has, by the profession’s own majority view, taken on a risk of error that the second edge requires them to manage: reading what the machine wrote before it enters the record. The same shape appears wherever adoption has run ahead of checking; the courts’ sanctions against lawyers who filed invented citations are the legal profession’s version, and how will AI change law collects them.

The duty to use travels with the duty to check#

Here the legal analysis meets the evidence this estate exists to hold. Checking an AI’s output requires the capability the tool displaces. A clinician who reads a scribe’s note for errors needs the documentation judgement that writing notes built. A lawyer who checks a citation needs to know the law well enough to notice the case does not exist. The measured risk is that routine use erodes that capability: the endoscopists on what is deskilling lost six percentage points of unassisted detection within months of assisted practice, and the expertise reversal effect describes how support that helps a novice can hold back an expert. The invisible work of oversight sets out why the checking rarely appears in anyone’s job description or time allocation. Put the statement beside the evidence and the legal position becomes a capability position. A professional is expected to use the tool and expected to be able to catch its errors; the second expectation is only meetable if the profession arranges to keep the skill that the first expectation stops exercising. What a person must remain able to do is now a fact a court may ask about.

Records are the other half of the case#

Cooley’s account records the statement’s view that causation will be the contested ground, because of “gaps in the evidence that arise because material has been destroyed, tampered with or simply not gathered”, and that courts might apply presumptions against a defendant whose “failure to record information that it reasonably should have” obscures what happened. For a professional this is concrete. Which tool was used, on what version, with what prompt, what it produced, what the professional changed and why: a firm that cannot answer those questions may find the gap counted against it. TLT’s summary of the statement puts record-keeping first in the list of what is “likely to prove decisive on questions of liability”, alongside human oversight, due diligence and transparency. A record of the check is also the only evidence that the check happened.

What a firm decides before the standard decides for it#

A profession or a firm can wait for the courts to say what a competent member would have done, or it can write it down. The list is short and it is the one this estate keeps arriving at, set out in Rules Before Tools: the tasks for which the firm expects its professionals to use AI, because the evidence and peer practice support it; the tasks for which it does not permit it, because the output cannot be checked at the point of use; who checks each class of output and how long that takes, in the terms of decision rights; what each grade must remain able to do unaided, and how the firm would know if that ability had gone; and what is recorded. A firm that has written that list has defined its own standard of care in advance and has the records to show it was followed. A firm that has not will have the standard defined for it by the survey a claimant’s counsel finds.

What this does not show#

It does not show that any professional has been found liable for not using AI; the statement is an analysis of what the law is likely to do, not a judgment, and it was read here through five law firms’ and one trade title’s summaries rather than at source. It does not show what “a competent member of the profession” uses in any profession other than UK general practice, and there the survey is a self-selected sample skewed young and male. It does not show that GPs are failing to check scribe output; the survey measured attitudes to risk, not checking behaviour. The deskilling evidence comes from endoscopy and from education research, and how far it transfers to documentation or legal drafting is not measured. What the record supports is narrower: the standard of care now has an edge on each side, both edges are set by what the profession does, and the capability to check is the one thing on which both depend.